omniglot: Same alphabet appears in "train" and "test"

[jvlmdr]

In data/omniglot/splits/vinyals, one alphabet (Gurmukhi) appears in both train.txt and test.txt.

I realise that this may not be a mistake if it corresponds to the original splits used. However, it should probably be fixed moving forward.

Note that none of the characters appear in both train and test sets, just that some characters from that alphabet are in train and some are in test.

(Hopefully it didn't affect any of the results since there are only 4 characters from this alphabet in the test set -- not enough to do 5-way classification.)